@tangle-network/agent-eval 0.200.0 → 0.201.0

package/dist/{integrity-BFP4kjGj.js → integrity-BvnVARVl.js}

@@ -82,37 +82,38 @@ async function assertRunCaptured(store, runId, expectations = {}) {
 		covered: 0,
 		total: llmSpans.length
 	};
-	if (expectations.rawSink) if (!expectations.rawSink.list) issues.push({
-		code: "no_raw_sink",
-		message: "Provided rawSink does not implement list(); cannot verify capture."
-	});
-	else {
-		const events = await expectations.rawSink.list({ runId });
-		rawEventCount = events.length;
-		const rawMin = expectations.rawProviderEventsMin ?? 1;
-		if (rawEventCount < rawMin) issues.push({
-			code: "missing_raw_events",
-			message: `Expected ≥ ${rawMin} raw provider events, found ${rawEventCount}.`,
-			detail: {
-				expected: rawMin,
-				found: rawEventCount
-			}
+	if (expectations.rawSink) {
+		if (!expectations.rawSink.list) issues.push({
+			code: "no_raw_sink",
+			message: "Provided rawSink does not implement list(); cannot verify capture."
 		});
-		if (expectations.requireRawCoverageOfLlmSpans) {
-			const requestEventsBySpan = new Set(events.filter((e) => e.direction === "request" && e.spanId).map((e) => e.spanId));
-			const orphaned = llmSpans.filter((s) => !requestEventsBySpan.has(s.spanId));
-			coverage = {
-				covered: llmSpans.length - orphaned.length,
-				total: llmSpans.length
-			};
-			if (orphaned.length > 0) issues.push({
-				code: "orphan_llm_span",
-				message: `${orphaned.length} LLM span(s) have no matching raw provider request event.`,
-				detail: { orphanedSpanIds: orphaned.map((s) => s.spanId) }
+		else {
+			const events = await expectations.rawSink.list({ runId });
+			rawEventCount = events.length;
+			const rawMin = expectations.rawProviderEventsMin ?? 1;
+			if (rawEventCount < rawMin) issues.push({
+				code: "missing_raw_events",
+				message: `Expected ≥ ${rawMin} raw provider events, found ${rawEventCount}.`,
+				detail: {
+					expected: rawMin,
+					found: rawEventCount
+				}
 			});
+			if (expectations.requireRawCoverageOfLlmSpans) {
+				const requestEventsBySpan = new Set(events.filter((e) => e.direction === "request" && e.spanId).map((e) => e.spanId));
+				const orphaned = llmSpans.filter((s) => !requestEventsBySpan.has(s.spanId));
+				coverage = {
+					covered: llmSpans.length - orphaned.length,
+					total: llmSpans.length
+				};
+				if (orphaned.length > 0) issues.push({
+					code: "orphan_llm_span",
+					message: `${orphaned.length} LLM span(s) have no matching raw provider request event.`,
+					detail: { orphanedSpanIds: orphaned.map((s) => s.spanId) }
+				});
+			}
 		}
-	}
-	else if (expectations.requireRawCoverageOfLlmSpans || expectations.rawProviderEventsMin) issues.push({
+	} else if (expectations.requireRawCoverageOfLlmSpans || expectations.rawProviderEventsMin) issues.push({
 		code: "no_raw_sink",
 		message: "Raw coverage required but no rawSink supplied to the integrity check."
 	});
@@ -143,4 +144,4 @@ function throwIfRunIncomplete(report) {
 //#endregion
 export { assertRunCaptured as n, throwIfRunIncomplete as r, RunIntegrityError as t };
 
-//# sourceMappingURL=integrity-BFP4kjGj.js.map
+//# sourceMappingURL=integrity-BvnVARVl.js.map

package/dist/{integrity-BFP4kjGj.js.map → integrity-BvnVARVl.js.map}

@@ -1 +1 @@
-{"version":3,"file":"integrity-BFP4kjGj.js","names":[],"sources":["../src/trace/integrity.ts"],"sourcesContent":["/**\n * Run-completion integrity check — at end of run, verify the expected event\n * types were actually captured. The point is the launch-review failure mode:\n * a run *appears* successful but the raw provider events were never written,\n * so a downstream reviewer can't reconstruct what happened.\n *\n * Pattern:\n *\n *   const report = await assertRunCaptured(store, runId, {\n *     llmSpansMin: 1,\n *     judgeSpansMin: 1,\n *     rawSink: providerSink,                  // must have ≥ 1 event for this run\n *     requireRawCoverageOfLlmSpans: true,     // every llm span has matching raw events\n *   })\n *   if (!report.ok) throwIfRunIncomplete(report)  // or mark run failed and continue\n *\n * The function is read-only on the store and returns a structured report;\n * the caller chooses the failure mode (throw, mark run failed, log warning).\n * `throwIfRunIncomplete` is the convenient strict mode.\n */\n\nimport { CaptureIntegrityError } from '../errors'\nimport type { RawProviderSink } from './raw-provider-sink'\nimport type { TraceStore } from './store'\n\nexport interface RunIntegrityExpectations {\n  /** Minimum LLM span count. Default 0 (no requirement). */\n  llmSpansMin?: number\n  /** Minimum judge span count. Default 0. */\n  judgeSpansMin?: number\n  /** Minimum tool span count. Default 0. */\n  toolSpansMin?: number\n  /**\n   * Raw provider sink to consult for capture verification. When present,\n   * the check requires at least one raw event for the run.\n   */\n  rawSink?: RawProviderSink\n  /** Minimum raw provider event count. Default 0; ignored when `rawSink` absent. */\n  rawProviderEventsMin?: number\n  /**\n   * Every LLM span must have at least one matching raw `request` event\n   * (matched by spanId). Catches the common bug where the structured span\n   * was emitted but the raw HTTP capture was wired to a different sink.\n   */\n  requireRawCoverageOfLlmSpans?: boolean\n  /** Run outcome must be set (not null/undefined). Default false. */\n  requireOutcome?: boolean\n}\n\nexport type RunIntegrityIssueCode =\n  | 'no_run'\n  | 'missing_llm_spans'\n  | 'missing_judge_spans'\n  | 'missing_tool_spans'\n  | 'missing_raw_events'\n  | 'no_raw_sink'\n  | 'orphan_llm_span'\n  | 'missing_outcome'\n  | 'dropped_writes'\n\nexport interface RunIntegrityIssue {\n  code: RunIntegrityIssueCode\n  message: string\n  detail?: Record<string, unknown>\n}\n\nexport interface RunIntegrityReport {\n  ok: boolean\n  runId: string\n  llmSpanCount: number\n  judgeSpanCount: number\n  toolSpanCount: number\n  rawProviderEventCount: number\n  /**\n   * Coverage of LLM spans by raw provider events keyed on spanId.\n   * `total` is the number of LLM spans; `covered` is the count with at\n   * least one matching `request` raw event.\n   */\n  rawSpanCoverage: { covered: number; total: number }\n  issues: RunIntegrityIssue[]\n}\n\nexport class RunIntegrityError extends CaptureIntegrityError {\n  constructor(public readonly report: RunIntegrityReport) {\n    super(\n      `Run ${report.runId} failed integrity check: ${report.issues.map((i) => i.code).join(', ')}`,\n    )\n  }\n}\n\nexport async function assertRunCaptured(\n  store: TraceStore,\n  runId: string,\n  expectations: RunIntegrityExpectations = {},\n): Promise<RunIntegrityReport> {\n  const issues: RunIntegrityIssue[] = []\n  const run = await store.getRun(runId)\n  if (!run) {\n    return {\n      ok: false,\n      runId,\n      llmSpanCount: 0,\n      judgeSpanCount: 0,\n      toolSpanCount: 0,\n      rawProviderEventCount: 0,\n      rawSpanCoverage: { covered: 0, total: 0 },\n      issues: [{ code: 'no_run', message: `Run ${runId} not found in store.` }],\n    }\n  }\n\n  const spans = await store.spans({ runId })\n  const llmSpans = spans.filter((s) => s.kind === 'llm')\n  const judgeSpans = spans.filter((s) => s.kind === 'judge')\n  const toolSpans = spans.filter((s) => s.kind === 'tool')\n\n  const llmMin = expectations.llmSpansMin ?? 0\n  const judgeMin = expectations.judgeSpansMin ?? 0\n  const toolMin = expectations.toolSpansMin ?? 0\n\n  if (llmSpans.length < llmMin) {\n    issues.push({\n      code: 'missing_llm_spans',\n      message: `Expected ≥ ${llmMin} LLM spans, found ${llmSpans.length}.`,\n      detail: { expected: llmMin, found: llmSpans.length },\n    })\n  }\n  if (judgeSpans.length < judgeMin) {\n    issues.push({\n      code: 'missing_judge_spans',\n      message: `Expected ≥ ${judgeMin} judge spans, found ${judgeSpans.length}.`,\n      detail: { expected: judgeMin, found: judgeSpans.length },\n    })\n  }\n  if (toolSpans.length < toolMin) {\n    issues.push({\n      code: 'missing_tool_spans',\n      message: `Expected ≥ ${toolMin} tool spans, found ${toolSpans.length}.`,\n      detail: { expected: toolMin, found: toolSpans.length },\n    })\n  }\n\n  let rawEventCount = 0\n  let coverage = { covered: 0, total: llmSpans.length }\n\n  if (expectations.rawSink) {\n    if (!expectations.rawSink.list) {\n      issues.push({\n        code: 'no_raw_sink',\n        message: 'Provided rawSink does not implement list(); cannot verify capture.',\n      })\n    } else {\n      const events = await expectations.rawSink.list({ runId })\n      rawEventCount = events.length\n      const rawMin = expectations.rawProviderEventsMin ?? 1\n      if (rawEventCount < rawMin) {\n        issues.push({\n          code: 'missing_raw_events',\n          message: `Expected ≥ ${rawMin} raw provider events, found ${rawEventCount}.`,\n          detail: { expected: rawMin, found: rawEventCount },\n        })\n      }\n      if (expectations.requireRawCoverageOfLlmSpans) {\n        const requestEventsBySpan = new Set(\n          events.filter((e) => e.direction === 'request' && e.spanId).map((e) => e.spanId!),\n        )\n        const orphaned = llmSpans.filter((s) => !requestEventsBySpan.has(s.spanId))\n        coverage = { covered: llmSpans.length - orphaned.length, total: llmSpans.length }\n        if (orphaned.length > 0) {\n          issues.push({\n            code: 'orphan_llm_span',\n            message: `${orphaned.length} LLM span(s) have no matching raw provider request event.`,\n            detail: { orphanedSpanIds: orphaned.map((s) => s.spanId) },\n          })\n        }\n      }\n    }\n  } else if (expectations.requireRawCoverageOfLlmSpans || expectations.rawProviderEventsMin) {\n    issues.push({\n      code: 'no_raw_sink',\n      message: 'Raw coverage required but no rawSink supplied to the integrity check.',\n    })\n  }\n\n  // The producer's own count of failed store writes. Every threshold above\n  // counts what reached the store; this is the record of what did not.\n  if (run.capture && run.capture.dropped > 0) {\n    issues.push({\n      code: 'dropped_writes',\n      message: `Run ${runId} dropped ${run.capture.dropped} of ${run.capture.written + run.capture.dropped} store writes; its records are incomplete.`,\n      detail: { ...run.capture },\n    })\n  }\n\n  if (expectations.requireOutcome && (run.outcome === undefined || run.outcome === null)) {\n    issues.push({\n      code: 'missing_outcome',\n      message: `Run ${runId} has no outcome recorded.`,\n    })\n  }\n\n  return {\n    ok: issues.length === 0,\n    runId,\n    llmSpanCount: llmSpans.length,\n    judgeSpanCount: judgeSpans.length,\n    toolSpanCount: toolSpans.length,\n    rawProviderEventCount: rawEventCount,\n    rawSpanCoverage: coverage,\n    issues,\n  }\n}\n\n/** Strict mode: throws `RunIntegrityError` when the report isn't ok. */\nexport function throwIfRunIncomplete(report: RunIntegrityReport): void {\n  if (!report.ok) throw new RunIntegrityError(report)\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AAkFA,IAAa,oBAAb,cAAuC,sBAAsB;CAC/B;CAA5B,YAAY,QAA4C;EACtD,MACE,OAAO,OAAO,MAAM,2BAA2B,OAAO,OAAO,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,IAAI,GAC3F;EAH0B,KAAA,SAAA;CAI5B;AACF;AAEA,eAAsB,kBACpB,OACA,OACA,eAAyC,CAAC,GACb;CAC7B,MAAM,SAA8B,CAAC;CACrC,MAAM,MAAM,MAAM,MAAM,OAAO,KAAK;CACpC,IAAI,CAAC,KACH,OAAO;EACL,IAAI;EACJ;EACA,cAAc;EACd,gBAAgB;EAChB,eAAe;EACf,uBAAuB;EACvB,iBAAiB;GAAE,SAAS;GAAG,OAAO;EAAE;EACxC,QAAQ,CAAC;GAAE,MAAM;GAAU,SAAS,OAAO,MAAM;EAAsB,CAAC;CAC1E;CAGF,MAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,MAAM,CAAC;CACzC,MAAM,WAAW,MAAM,QAAQ,MAAM,EAAE,SAAS,KAAK;CACrD,MAAM,aAAa,MAAM,QAAQ,MAAM,EAAE,SAAS,OAAO;CACzD,MAAM,YAAY,MAAM,QAAQ,MAAM,EAAE,SAAS,MAAM;CAEvD,MAAM,SAAS,aAAa,eAAe;CAC3C,MAAM,WAAW,aAAa,iBAAiB;CAC/C,MAAM,UAAU,aAAa,gBAAgB;CAE7C,IAAI,SAAS,SAAS,QACpB,OAAO,KAAK;EACV,MAAM;EACN,SAAS,cAAc,OAAO,oBAAoB,SAAS,OAAO;EAClE,QAAQ;GAAE,UAAU;GAAQ,OAAO,SAAS;EAAO;CACrD,CAAC;CAEH,IAAI,WAAW,SAAS,UACtB,OAAO,KAAK;EACV,MAAM;EACN,SAAS,cAAc,SAAS,sBAAsB,WAAW,OAAO;EACxE,QAAQ;GAAE,UAAU;GAAU,OAAO,WAAW;EAAO;CACzD,CAAC;CAEH,IAAI,UAAU,SAAS,SACrB,OAAO,KAAK;EACV,MAAM;EACN,SAAS,cAAc,QAAQ,qBAAqB,UAAU,OAAO;EACrE,QAAQ;GAAE,UAAU;GAAS,OAAO,UAAU;EAAO;CACvD,CAAC;CAGH,IAAI,gBAAgB;CACpB,IAAI,WAAW;EAAE,SAAS;EAAG,OAAO,SAAS;CAAO;CAEpD,IAAI,aAAa,SACf,IAAI,CAAC,aAAa,QAAQ,MACxB,OAAO,KAAK;EACV,MAAM;EACN,SAAS;CACX,CAAC;MACI;EACL,MAAM,SAAS,MAAM,aAAa,QAAQ,KAAK,EAAE,MAAM,CAAC;EACxD,gBAAgB,OAAO;EACvB,MAAM,SAAS,aAAa,wBAAwB;EACpD,IAAI,gBAAgB,QAClB,OAAO,KAAK;GACV,MAAM;GACN,SAAS,cAAc,OAAO,8BAA8B,cAAc;GAC1E,QAAQ;IAAE,UAAU;IAAQ,OAAO;GAAc;EACnD,CAAC;EAEH,IAAI,aAAa,8BAA8B;GAC7C,MAAM,sBAAsB,IAAI,IAC9B,OAAO,QAAQ,MAAM,EAAE,cAAc,aAAa,EAAE,MAAM,CAAC,CAAC,KAAK,MAAM,EAAE,MAAO,CAClF;GACA,MAAM,WAAW,SAAS,QAAQ,MAAM,CAAC,oBAAoB,IAAI,EAAE,MAAM,CAAC;GAC1E,WAAW;IAAE,SAAS,SAAS,SAAS,SAAS;IAAQ,OAAO,SAAS;GAAO;GAChF,IAAI,SAAS,SAAS,GACpB,OAAO,KAAK;IACV,MAAM;IACN,SAAS,GAAG,SAAS,OAAO;IAC5B,QAAQ,EAAE,iBAAiB,SAAS,KAAK,MAAM,EAAE,MAAM,EAAE;GAC3D,CAAC;EAEL;CACF;MACK,IAAI,aAAa,gCAAgC,aAAa,sBACnE,OAAO,KAAK;EACV,MAAM;EACN,SAAS;CACX,CAAC;CAKH,IAAI,IAAI,WAAW,IAAI,QAAQ,UAAU,GACvC,OAAO,KAAK;EACV,MAAM;EACN,SAAS,OAAO,MAAM,WAAW,IAAI,QAAQ,QAAQ,MAAM,IAAI,QAAQ,UAAU,IAAI,QAAQ,QAAQ;EACrG,QAAQ,EAAE,GAAG,IAAI,QAAQ;CAC3B,CAAC;CAGH,IAAI,aAAa,mBAAmB,IAAI,YAAY,KAAA,KAAa,IAAI,YAAY,OAC/E,OAAO,KAAK;EACV,MAAM;EACN,SAAS,OAAO,MAAM;CACxB,CAAC;CAGH,OAAO;EACL,IAAI,OAAO,WAAW;EACtB;EACA,cAAc,SAAS;EACvB,gBAAgB,WAAW;EAC3B,eAAe,UAAU;EACzB,uBAAuB;EACvB,iBAAiB;EACjB;CACF;AACF;;AAGA,SAAgB,qBAAqB,QAAkC;CACrE,IAAI,CAAC,OAAO,IAAI,MAAM,IAAI,kBAAkB,MAAM;AACpD"}
+{"version":3,"file":"integrity-BvnVARVl.js","names":[],"sources":["../src/trace/integrity.ts"],"sourcesContent":["/**\n * Run-completion integrity check — at end of run, verify the expected event\n * types were actually captured. The point is the launch-review failure mode:\n * a run *appears* successful but the raw provider events were never written,\n * so a downstream reviewer can't reconstruct what happened.\n *\n * Pattern:\n *\n *   const report = await assertRunCaptured(store, runId, {\n *     llmSpansMin: 1,\n *     judgeSpansMin: 1,\n *     rawSink: providerSink,                  // must have ≥ 1 event for this run\n *     requireRawCoverageOfLlmSpans: true,     // every llm span has matching raw events\n *   })\n *   if (!report.ok) throwIfRunIncomplete(report)  // or mark run failed and continue\n *\n * The function is read-only on the store and returns a structured report;\n * the caller chooses the failure mode (throw, mark run failed, log warning).\n * `throwIfRunIncomplete` is the convenient strict mode.\n */\n\nimport { CaptureIntegrityError } from '../errors'\nimport type { RawProviderSink } from './raw-provider-sink'\nimport type { TraceStore } from './store'\n\nexport interface RunIntegrityExpectations {\n  /** Minimum LLM span count. Default 0 (no requirement). */\n  llmSpansMin?: number\n  /** Minimum judge span count. Default 0. */\n  judgeSpansMin?: number\n  /** Minimum tool span count. Default 0. */\n  toolSpansMin?: number\n  /**\n   * Raw provider sink to consult for capture verification. When present,\n   * the check requires at least one raw event for the run.\n   */\n  rawSink?: RawProviderSink\n  /** Minimum raw provider event count. Default 0; ignored when `rawSink` absent. */\n  rawProviderEventsMin?: number\n  /**\n   * Every LLM span must have at least one matching raw `request` event\n   * (matched by spanId). Catches the common bug where the structured span\n   * was emitted but the raw HTTP capture was wired to a different sink.\n   */\n  requireRawCoverageOfLlmSpans?: boolean\n  /** Run outcome must be set (not null/undefined). Default false. */\n  requireOutcome?: boolean\n}\n\nexport type RunIntegrityIssueCode =\n  | 'no_run'\n  | 'missing_llm_spans'\n  | 'missing_judge_spans'\n  | 'missing_tool_spans'\n  | 'missing_raw_events'\n  | 'no_raw_sink'\n  | 'orphan_llm_span'\n  | 'missing_outcome'\n  | 'dropped_writes'\n\nexport interface RunIntegrityIssue {\n  code: RunIntegrityIssueCode\n  message: string\n  detail?: Record<string, unknown>\n}\n\nexport interface RunIntegrityReport {\n  ok: boolean\n  runId: string\n  llmSpanCount: number\n  judgeSpanCount: number\n  toolSpanCount: number\n  rawProviderEventCount: number\n  /**\n   * Coverage of LLM spans by raw provider events keyed on spanId.\n   * `total` is the number of LLM spans; `covered` is the count with at\n   * least one matching `request` raw event.\n   */\n  rawSpanCoverage: { covered: number; total: number }\n  issues: RunIntegrityIssue[]\n}\n\nexport class RunIntegrityError extends CaptureIntegrityError {\n  constructor(public readonly report: RunIntegrityReport) {\n    super(\n      `Run ${report.runId} failed integrity check: ${report.issues.map((i) => i.code).join(', ')}`,\n    )\n  }\n}\n\nexport async function assertRunCaptured(\n  store: TraceStore,\n  runId: string,\n  expectations: RunIntegrityExpectations = {},\n): Promise<RunIntegrityReport> {\n  const issues: RunIntegrityIssue[] = []\n  const run = await store.getRun(runId)\n  if (!run) {\n    return {\n      ok: false,\n      runId,\n      llmSpanCount: 0,\n      judgeSpanCount: 0,\n      toolSpanCount: 0,\n      rawProviderEventCount: 0,\n      rawSpanCoverage: { covered: 0, total: 0 },\n      issues: [{ code: 'no_run', message: `Run ${runId} not found in store.` }],\n    }\n  }\n\n  const spans = await store.spans({ runId })\n  const llmSpans = spans.filter((s) => s.kind === 'llm')\n  const judgeSpans = spans.filter((s) => s.kind === 'judge')\n  const toolSpans = spans.filter((s) => s.kind === 'tool')\n\n  const llmMin = expectations.llmSpansMin ?? 0\n  const judgeMin = expectations.judgeSpansMin ?? 0\n  const toolMin = expectations.toolSpansMin ?? 0\n\n  if (llmSpans.length < llmMin) {\n    issues.push({\n      code: 'missing_llm_spans',\n      message: `Expected ≥ ${llmMin} LLM spans, found ${llmSpans.length}.`,\n      detail: { expected: llmMin, found: llmSpans.length },\n    })\n  }\n  if (judgeSpans.length < judgeMin) {\n    issues.push({\n      code: 'missing_judge_spans',\n      message: `Expected ≥ ${judgeMin} judge spans, found ${judgeSpans.length}.`,\n      detail: { expected: judgeMin, found: judgeSpans.length },\n    })\n  }\n  if (toolSpans.length < toolMin) {\n    issues.push({\n      code: 'missing_tool_spans',\n      message: `Expected ≥ ${toolMin} tool spans, found ${toolSpans.length}.`,\n      detail: { expected: toolMin, found: toolSpans.length },\n    })\n  }\n\n  let rawEventCount = 0\n  let coverage = { covered: 0, total: llmSpans.length }\n\n  if (expectations.rawSink) {\n    if (!expectations.rawSink.list) {\n      issues.push({\n        code: 'no_raw_sink',\n        message: 'Provided rawSink does not implement list(); cannot verify capture.',\n      })\n    } else {\n      const events = await expectations.rawSink.list({ runId })\n      rawEventCount = events.length\n      const rawMin = expectations.rawProviderEventsMin ?? 1\n      if (rawEventCount < rawMin) {\n        issues.push({\n          code: 'missing_raw_events',\n          message: `Expected ≥ ${rawMin} raw provider events, found ${rawEventCount}.`,\n          detail: { expected: rawMin, found: rawEventCount },\n        })\n      }\n      if (expectations.requireRawCoverageOfLlmSpans) {\n        const requestEventsBySpan = new Set(\n          events.filter((e) => e.direction === 'request' && e.spanId).map((e) => e.spanId!),\n        )\n        const orphaned = llmSpans.filter((s) => !requestEventsBySpan.has(s.spanId))\n        coverage = { covered: llmSpans.length - orphaned.length, total: llmSpans.length }\n        if (orphaned.length > 0) {\n          issues.push({\n            code: 'orphan_llm_span',\n            message: `${orphaned.length} LLM span(s) have no matching raw provider request event.`,\n            detail: { orphanedSpanIds: orphaned.map((s) => s.spanId) },\n          })\n        }\n      }\n    }\n  } else if (expectations.requireRawCoverageOfLlmSpans || expectations.rawProviderEventsMin) {\n    issues.push({\n      code: 'no_raw_sink',\n      message: 'Raw coverage required but no rawSink supplied to the integrity check.',\n    })\n  }\n\n  // The producer's own count of failed store writes. Every threshold above\n  // counts what reached the store; this is the record of what did not.\n  if (run.capture && run.capture.dropped > 0) {\n    issues.push({\n      code: 'dropped_writes',\n      message: `Run ${runId} dropped ${run.capture.dropped} of ${run.capture.written + run.capture.dropped} store writes; its records are incomplete.`,\n      detail: { ...run.capture },\n    })\n  }\n\n  if (expectations.requireOutcome && (run.outcome === undefined || run.outcome === null)) {\n    issues.push({\n      code: 'missing_outcome',\n      message: `Run ${runId} has no outcome recorded.`,\n    })\n  }\n\n  return {\n    ok: issues.length === 0,\n    runId,\n    llmSpanCount: llmSpans.length,\n    judgeSpanCount: judgeSpans.length,\n    toolSpanCount: toolSpans.length,\n    rawProviderEventCount: rawEventCount,\n    rawSpanCoverage: coverage,\n    issues,\n  }\n}\n\n/** Strict mode: throws `RunIntegrityError` when the report isn't ok. */\nexport function throwIfRunIncomplete(report: RunIntegrityReport): void {\n  if (!report.ok) throw new RunIntegrityError(report)\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AAkFA,IAAa,oBAAb,cAAuC,sBAAsB;CAC/B;CAA5B,YAAY,QAA4C;EACtD,MACE,OAAO,OAAO,MAAM,2BAA2B,OAAO,OAAO,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,IAAI,GAC3F;EAH0B,KAAA,SAAA;CAI5B;AACF;AAEA,eAAsB,kBACpB,OACA,OACA,eAAyC,CAAC,GACb;CAC7B,MAAM,SAA8B,CAAC;CACrC,MAAM,MAAM,MAAM,MAAM,OAAO,KAAK;CACpC,IAAI,CAAC,KACH,OAAO;EACL,IAAI;EACJ;EACA,cAAc;EACd,gBAAgB;EAChB,eAAe;EACf,uBAAuB;EACvB,iBAAiB;GAAE,SAAS;GAAG,OAAO;EAAE;EACxC,QAAQ,CAAC;GAAE,MAAM;GAAU,SAAS,OAAO,MAAM;EAAsB,CAAC;CAC1E;CAGF,MAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,MAAM,CAAC;CACzC,MAAM,WAAW,MAAM,QAAQ,MAAM,EAAE,SAAS,KAAK;CACrD,MAAM,aAAa,MAAM,QAAQ,MAAM,EAAE,SAAS,OAAO;CACzD,MAAM,YAAY,MAAM,QAAQ,MAAM,EAAE,SAAS,MAAM;CAEvD,MAAM,SAAS,aAAa,eAAe;CAC3C,MAAM,WAAW,aAAa,iBAAiB;CAC/C,MAAM,UAAU,aAAa,gBAAgB;CAE7C,IAAI,SAAS,SAAS,QACpB,OAAO,KAAK;EACV,MAAM;EACN,SAAS,cAAc,OAAO,oBAAoB,SAAS,OAAO;EAClE,QAAQ;GAAE,UAAU;GAAQ,OAAO,SAAS;EAAO;CACrD,CAAC;CAEH,IAAI,WAAW,SAAS,UACtB,OAAO,KAAK;EACV,MAAM;EACN,SAAS,cAAc,SAAS,sBAAsB,WAAW,OAAO;EACxE,QAAQ;GAAE,UAAU;GAAU,OAAO,WAAW;EAAO;CACzD,CAAC;CAEH,IAAI,UAAU,SAAS,SACrB,OAAO,KAAK;EACV,MAAM;EACN,SAAS,cAAc,QAAQ,qBAAqB,UAAU,OAAO;EACrE,QAAQ;GAAE,UAAU;GAAS,OAAO,UAAU;EAAO;CACvD,CAAC;CAGH,IAAI,gBAAgB;CACpB,IAAI,WAAW;EAAE,SAAS;EAAG,OAAO,SAAS;CAAO;CAEpD,IAAI,aAAa,SAAS;EACxB,IAAI,CAAC,aAAa,QAAQ,MACxB,OAAO,KAAK;GACV,MAAM;GACN,SAAS;EACX,CAAC;OACI;GACL,MAAM,SAAS,MAAM,aAAa,QAAQ,KAAK,EAAE,MAAM,CAAC;GACxD,gBAAgB,OAAO;GACvB,MAAM,SAAS,aAAa,wBAAwB;GACpD,IAAI,gBAAgB,QAClB,OAAO,KAAK;IACV,MAAM;IACN,SAAS,cAAc,OAAO,8BAA8B,cAAc;IAC1E,QAAQ;KAAE,UAAU;KAAQ,OAAO;IAAc;GACnD,CAAC;GAEH,IAAI,aAAa,8BAA8B;IAC7C,MAAM,sBAAsB,IAAI,IAC9B,OAAO,QAAQ,MAAM,EAAE,cAAc,aAAa,EAAE,MAAM,CAAC,CAAC,KAAK,MAAM,EAAE,MAAO,CAClF;IACA,MAAM,WAAW,SAAS,QAAQ,MAAM,CAAC,oBAAoB,IAAI,EAAE,MAAM,CAAC;IAC1E,WAAW;KAAE,SAAS,SAAS,SAAS,SAAS;KAAQ,OAAO,SAAS;IAAO;IAChF,IAAI,SAAS,SAAS,GACpB,OAAO,KAAK;KACV,MAAM;KACN,SAAS,GAAG,SAAS,OAAO;KAC5B,QAAQ,EAAE,iBAAiB,SAAS,KAAK,MAAM,EAAE,MAAM,EAAE;IAC3D,CAAC;GAEL;EACF;CACF,OAAO,IAAI,aAAa,gCAAgC,aAAa,sBACnE,OAAO,KAAK;EACV,MAAM;EACN,SAAS;CACX,CAAC;CAKH,IAAI,IAAI,WAAW,IAAI,QAAQ,UAAU,GACvC,OAAO,KAAK;EACV,MAAM;EACN,SAAS,OAAO,MAAM,WAAW,IAAI,QAAQ,QAAQ,MAAM,IAAI,QAAQ,UAAU,IAAI,QAAQ,QAAQ;EACrG,QAAQ,EAAE,GAAG,IAAI,QAAQ;CAC3B,CAAC;CAGH,IAAI,aAAa,mBAAmB,IAAI,YAAY,KAAA,KAAa,IAAI,YAAY,OAC/E,OAAO,KAAK;EACV,MAAM;EACN,SAAS,OAAO,MAAM;CACxB,CAAC;CAGH,OAAO;EACL,IAAI,OAAO,WAAW;EACtB;EACA,cAAc,SAAS;EACvB,gBAAgB,WAAW;EAC3B,eAAe,UAAU;EACzB,uBAAuB;EACvB,iBAAiB;EACjB;CACF;AACF;;AAGA,SAAgB,qBAAqB,QAAkC;CACrE,IAAI,CAAC,OAAO,IAAI,MAAM,IAAI,kBAAkB,MAAM;AACpD"}

package/dist/{integrity-BvHANvnp.js → integrity-uwK7XC-8.js}

@@ -1,5 +1,5 @@
 import { c as unscreenedRewardFields } from "./reward-nw2xZGZG.js";
-import { f as validateRolloutLine, r as ROLLOUT_SCHEMA, u as isRolloutLine } from "./schema-C1aaAxTf.js";
+import { f as validateRolloutLine, r as ROLLOUT_SCHEMA, u as isRolloutLine } from "./schema-DAMp77zB.js";
 import { a as parseJson, f as isUnavailable, i as asRecord, r as readTerminalRecord, s as parseSupervisorTree } from "./terminal-record-BHocwFw1.js";
 //#region src/supervisor-run/integrity-issues.ts
 function evidence(path, value) {
@@ -1161,4 +1161,4 @@ function analyzeSupervisorRunIntegrity(input, options = {}) {
 //#endregion
 export { supervisorRunRolloutLines as n, SUPERVISOR_RUN_INTEGRITY_SCHEMA as r, analyzeSupervisorRunIntegrity as t };
 
-//# sourceMappingURL=integrity-BvHANvnp.js.map
+//# sourceMappingURL=integrity-uwK7XC-8.js.map